i watched severance tonight and the  first thing that popped into
my head when it was over was the realization that traditional row level locking could be paired with serializable snapshot isolation to increase write throughput in situations where high contention on a specific rows causes frequent rollbacks

Building @warpcast and @farcaster / dad / like to read, cook, ski, code

I have yet to see severance, but I don't follow how this improves throughput?

BDFL of Quilibrium, Eng @ Farcaster, ex-Coinbase, always opinionated, never hydrated

/quilibrium

it appears they already do this under the hood, but with a bit of nuance to it (Section 5.2)

it appears they already do this under the hood, but with a bit of nuance to it (Section 5.2) https://arxiv.org/pdf/1208.4179

thinking is that having concurrent statements you know would conflict sequence themselves with the row level lock rather is more efficient than having them all going through execute / rollback / retry until they all succeed

haven’t had a chance to test it out though